Wednesday 5 December 2012

Aggregating Everything - Map/Reduce and Camel?

If you are used to Map/Reduce you will be used to the idea of breaking tasks down into little chunks and then collecting the partial results together into some final format.

So, recently when I was parsing zillions of rows of data and aggregating related data into partial CSV files  and then aggregating the bits of partial of data to reports I thought - Aha! MapReduce.

For a whole bunch of good design decisions I was using Apache Camel - a neat pipelining tool which with a bit of help from ActiveMQ provides the sort of long running stability that I needed. Camel however does not do Map/Reduce, but it does have the Aggregator Integration pattern, which you can use to so a similar thing.

Image courtesy of Carlos Oliveira
Imagine you empty your jar of loose change on a table. You stack the nickles in stacks of ten coins, the dimes is stacks of ten coins and the quarters in stacks of ten coins. You add up all the 50 cents, $1s and $2.50s and you know how much you have. That's Map/Reduce.

Now, imagine you empty your jar of loose change into one of those coin counting machines in the Mall. Internally all the coins are sorted by falling through a hole which is either nickle, dime or quarter shaped and as they emerge from the other side they are counted*. That's aggregation Camel style.

I did hit a bit of a snag. I couldn't work out how to tell the Aggregator Integration pattern that there were no more files to come... Stop... Woaa-there... Desist!

It turns out that hidden away (in the middle of the docs) the File endpoint rather usefully sets a flag in the headers called CamelBatchComplete which is just what I was looking for:

Good luck fellow travelers.

* I have no idea how a coin counting machine works.